摘要
全景图拼接是将具有共同部分的多幅图像进行组合,实现一幅全景图的过程。针对基于传统SIFT(scaleinvariant feature transform)算法全景拼接中的特征点匹配计算消耗时间过长和存在冗余错误的不足提出了改进。其中,传统算法的特征点匹配计算是基于KD-tree算法的树结构,由近及远地逐个查找并计算特征点的匹配度;改进后的最近邻搜索算法(best-bin-first,BBF)是先根据每个特征点的多维度分量特性对其进行优先级排序,查询时总是从优先级高的开始,来提高匹配计算效率。冗余错误问题则是通过随机采样一致算法(RANSAC)的优化迭代计算错误概率,代替传统方法的阈值筛选法来减低错误匹配点的出现次数。实验中分别对简单纹理图像和复杂纹理图像进行了拼接实验并与原算法比较,证明本算法的拼接精度和时效性的提升。
Image mosaic aims to accomplish a complete image with stitching multiple images which overlap with useful image features.The paper introduces an improved algorithmwhich isto overcome the drawbacks of long-time-costingin feature point matching and high rate of wrong matching pointsbase on SIFT algorithm.In order to improve search efficiency when matching image features,the paper uses BBF（Best-Bin-First）search algorithm instead of KDtreealgorithm,because BBF matching method is carried out according to the priorities of SIFT features,which is established based on the significance of sub-feature within SIFT;While KD-tree matching algorithm is based on space position from nearness to farness.Furthermore,this paper applies modified RANSAC algorithm to purify wrong matched SIFT features by calculating their error probabilities;while the traditional way adopts a threshold to filter out these wrong matched feature points.Comparing test results with the original algorithm in terms of image mosaicking quality on simple texture images and complex texture images,the test results show that the proposed method can improve the accuracy and speed of image stitching.
出处
《电子测量技术》
2017年第7期90-94,99,共6页
Electronic Measurement Technology
基金
国家自然科学基金(61472196)
山东省自然科学基金(ZR2015FM012资助项目
